Introduction to the Special Issue: Case-Based Approaches to the Analysis of Quantitative Data

Abstract

Within the social sciences generally, the conventional approach to the analysis of survey data remains variable-based, employing some member of the regression family. Such methods address the effect of one or more supposedly “independent” variables on some outcome. Individual cases, the carriers of the variables, usually remain in the background, as do, often, underlying causal mechanisms and processes. It is variables that act, having their effects on a dependent outcome variable. In the typical multivariate study, the purpose is to report the net effect of each independent variable. The underlying mathematics is matrix algebra and the typical model additive. Notwithstanding the use of interaction terms and the development of multi-level modelling and related techniques, causal homogeneity is still often assumed across cases (an assumption whose realism was questioned by Ralph Turner as long ago as 1948). Over the past 30 years, a number of authors have published important critiques of the assumptions of this form of variable analysis. Abbott (2001), Byrne (2002), Freedman (1991), Lieberson (1985), Pawson (1989) and Ragin (1987, 2000, 2008), amongst others, have contributed much to our understanding of its limitations.